Study Design And Rationale

The following section examines the reasons for choosing a case-control design to understand the effect of conspicuity aids on bicycle crash risk and how the setting, participants, variables, sample size and statistical methods were

selected to increase the accuracy of the recording exposure and confounding variables and increase the validity of the results.

2.2.1 Intervention Studies

Cycling collisions are rare but can lead to significant morbidity and mortality. This rarity creates a considerable challenge for the empirical investigation of factors that may increase or decrease crash risk.

The “gold standard” for testing interventions in health research is widely considered to be the randomised controlled trial 217. A randomised controlled trial of the effect of conspicuity aids could be conducted to assess their effect on risk of collision. Such a study would have the advantage of allowing for the standardisation and unbiased allocation of the conspicuity intervention whilst ensuring a random distribution of potential confounding participant characteristics between the comparison groups. The large sample size, length of follow-up required, potentially high drop-out rates and the problem of ensuring compliance with the randomisation schedule mean that the resources and time required would be prohibitive.

Ethical considerations also restrict the applicability of randomised designs even in such ‘non-medical’ contexts as crash prevention research. The central ethical principle underlying the rationale for randomised controlled trials is that of clinical equipoise or the collective opinion of relevant professional groups, ethical committee members and others that there is insufficient evidence for the superiority of one intervention over another 218 p124. Conspicuity aids are mandatory in many industrial and transport settings. In the context of bicycle safety there is widespread advocacy of the use of conspicuity aids for cyclists and increasingly other vulnerable road users such as pedestrians amongst road safety and injury prevention professionals 202. This consensus view is based on a significant body of experimental research evidence of enhanced conspicuity if not of efficacy in preventing crashes 219. Randomly withholding a potentially beneficial intervention from a group of controls by randomisation is unlikely to be acceptable in ethical terms.

An alternative to random allocation are patient preference trials. Participants choose to adopt a given intervention but the assumption of equipoise remains at the level of the investigators who could be blinded for example. Such trials have

been suggested as a partial remedy to some ethical and practical difficulties with randomised designs as they can improve compliance and remove the ethical problems associated with random allocation 220. Despite addressing the issue of involuntary withholding of a potentially protective intervention, a preference trial would still need a large sample owing to the rarity of the crash outcome. Contamination of the intervention across groups might still occur further increasing the sample required. Any confounding association between conspicuity aid use and different safety behaviours would continue to threaten the validity of the results given that voluntary adopters are likely to differ in risk-taking behaviours leading to confounding of crash risk.

2.2.2 Observation Studies

Conspicuity aids are already used by some cyclists. Therefore the opportunity exists for an observational study of their efficacy. Observational studies have been conducted to investigate many injury prevention problems in an attempt to examine and quantify the potential causal or preventative role of various exposures on injury outcomes.

Prospective cohort study designs can provide an estimate of the incidence of a given outcome in a group of people observed over a period of time. They enable the calculation of relative risks or hazard ratios for comparison groups defined by exposure to a possible causal factor 221. The rarity of cycling crashes means that a cohort study of the effect of conspicuity aids would require a lengthy follow-up phase or large sample to deliver precise estimates of relative risk in a reasonable time period.

For example, assuming that there are 5000 regular cycle commuters in an available base population and a yearly rate of 150 adult cyclists injured in collision crashes a cohort study would need to recruit 3859 cyclists to detect a relative risk of 0.63 of a collision over a year for those using conspicuity aids (alpha = 0.05, beta 80%, 20% dropout rate; Stata Version 10). This is around 75% of the estimated number of regular commuter cyclists in Nottingham (National Transport Survey data). Contacting and recruiting such a large cohort would be prohibitive. Recent web-based designs have been reported but carry their own methodological problems despite the efficiency of online data collection tools 83 100.

2.2.3 Case-Control Studies

A more efficient observational design is that of the case-control study. The case- control design has often been used to investigate exposures in relatively rare events such as injury 222 and the techniques and analysis are now very refined 223

and widely applied 224. The method is resource efficient as cases are identified prospectively which drastically reduces the number of individuals who need to be followed-up to achieve sufficient statistical power. Controls are then identified from the same population which gave rise the cases. In the present study the outcome of interest is cycle crashes and cases are identified when they seek emergency department assessment and/or treatment. Controls are drawn from cyclists travelling in the same urban area at a similar time and for a similar purpose i.e. represent the same “base” population from which cases are derived. There are a number of potential limitations in the case-control method. Despite careful design, such studies cannot establish a causal link between the exposure and outcome of interest with any certainty 221. Rather, what is identified is an apparent association between an exposure variable and the outcome of interest. Once a case-control study has identified a potential effect of an exposure, further evidence is then required to confirm the findings. Even in the presence of a plausible causal mechanism other components are required such as a dose- response where increasing exposure is shown to be associated with greater risk of the disease. Study validity also requires the identification and minimisation of potential sources of confounding and bias.

The following section describes the methods adopted to examine the relationship between cyclists’ risk of collision crash and their use of conspicuity aids using a case-control design.